Includes learned parameters for:
(1) walk through scene (forcasting) only 36 weights because observations are not included
(2) walk through scene (smoothing) 40 weights because observation ARE included

* weights are in the same order as features in feature_maps
